Stable feature mining based object representation for tracking applications

Tracking the object with stable features is an important and challenging task in real scenes where the object appearance constantly changes or is disturbed by the background etc. In this paper, a novel frame work of object representation and tracking based on stable feature mining is presented. Firstly, the object region is adaptively detected and then the peak contour of V color component histogram in the object region is extracted to calculate candidate and region peaks which are used for acquiring the number of clusters and further classifying the object appearance. Secondly, the connection subregions belonging to every cluster are described with observation and increment models, and subregions association between the object template and the current observation is then utilized to mine stable subregion pairs and get feature change ratios. Finally, stable subregion displacements are weighted fused to locate the object in the current frame, and the object template is updated in terms of average increment variations. Experimental results show the excellent performance of the proposed algorithm in cluster number adaptivity, robustness in stable feature mining and accurateness in object tracking.